Math 215 HW #6 Solutions

1. Problem 3.1.14. Show that x − y is orthogonal to x + y if and only if kxk = kyk.

Proof. First, suppose x − y is orthogonal to x + y. Then

0 = hx − y, x + yi
= (x − y)T (x + y)
= xT x + xT y − yT x − yT x
= hx, xi + hx, yi − hy, xi−iy, yi
= hx, xi − hy, yi

since hx, yi = hy, xi. In other words,

0 = kxk2 − kyk2 ,

so kxk2 = kyk2 . Since the norm of a vector can never be negative, this implies that

kxk = kyk.

Thus, we see that if x − y is orthogonal to x + y, then kxk = kyk.

On the other hand, suppose kxk = kyk. Then

hx − y, x + yi = (x − y)T (x + y)
= xT x + xT y − yT x − yT x
= hx, xi + hx, yi − hy, xi−iy, yi
= hx, xi − hy, yi
= kxk2 − kyk2
= 0,

so we see that x − y is orthogonal to x + y.

We’ve seen that the implication goes both ways, so we conclude that x − y is orthogonal to
x + y if and only if kxk = kyk, as desired.

2. Problem 3.1.20. Let S be a subspace of Rn . Explain what (S⊥ )⊥ = S means and why it is
true.
Answer: First, (S⊥ )⊥ is the orthogonal complement of S⊥ , which is itself the orthogonal
complement of S, so (S⊥ )⊥ = S means that S is the orthogonal complement of its orthogonal
complement.
To show that it is true, we want to show that S is contained in (S⊥ )⊥ and, conversely, that
(S⊥ )⊥ is contained in S; if we can show both containments, then the only possible conclusion
is that (S⊥ )⊥ = S.
To show the first containment, suppose v ∈ S and w ∈ S⊥ . Then

hv, wi = 0

1
 by the definition of S⊥ . Thus, S is certainly contained in (S⊥ )⊥ (which consists of all vectors
in Rn which are orthogonal to S⊥ ).
To show the other containment, suppose v ∈ (S⊥ )⊥ (meaning that v is orthogonal to all
vectors in S⊥ ); then we want to show that v ∈ S. I’m sure there must be a better way to see
this, but here’s one that works. Let {u1 , . . . , up } be a basis for S and let {w1 , . . . , wq } be a
basis for S⊥ . If v ∈
/ S, then {u1 , . . . , up , v} is a linearly independent set. Since each vector
in that set is orthogonal to all of S⊥ , the set

{u1 , . . . , up , v, w1 , . . . , wq }

is linearly independent. Since there are p+q+1 vectors in this set, this means that p+q+1 ≤ n
or, equivalently, p + q ≤ n − 1. On the other hand, if A is the matrix whose ith row is uTi ,
then the row space of A is S and the nullspace of A is S⊥ . Since S is p-dimensional, the rank
of A is p, meaning that the dimension of nul(A) = S⊥ is q = n − p. Therefore,

p + q = p + (n − p) = n,

contradicting the fact that p + q ≤ n − 1. From this contradiction, then, we see that, if
v ∈ (S⊥ )⊥ , it must be the case that v ∈ S.

3. Problem 3.1.28. This is a system of equations Ax = bs with no solution:

x + 2y + 2z = 5
2x + 2y + 3z = 5
3x + 4y + 5z = 9.

Find numbers y1 , y2 , y3 to multiply the equations so they add to 0 = 1. You have found a
vector y in which subspace? The inner product is yT b is 1.
Answer: Let      
1 2 2 x 5
A =  2 2 3 , x =  y , and b =  5  .
3 4 5 z 9
Then the given system is equivalent to the matrix equation

Ax = b.

Let    
y1 1
y =  y2  =  1  .
y3 −1
Then multiplying the ith row in the system by yi and adding is equivalent to multiplying

2
 both sides of the matrix equation by yT . We see that
  
1 2 2 x
yT Ax = [1 1 − 1]  2 2 3   y 
3 4 5 z
 
x + 2y + 2z
= [1 1 − 1]  2x + 2y + 3z 
3x + 4y + 5z
= (x + 2y + 2z) + (2x + 2y + 3z) − (3x + 4y + 5z)
= 0.
On the other hand, 
5
yT b = [1 1 − 1]  5  = 5 + 5 − 9 = 1.
9
Therefore, yT Ax = yT b reduces to
0 = 1,
so we see that the system has no solution. Since
0 = yT Ax = hy, Axi,
we see that y is perpendicular to Ax no matter what x is. Therefore, the vector y is per-
pendicular to the column space of A. Since the orthogonal complement of col(A) is the left
nullspace of A, we see that y must be an element of the left nullspace of A.
4. Problem 3.1.36. Extend Problem 35 to a p-dimensional subspace V and a q-dimensional
subspace W of Rn . What inequality on p + q guarantees that V intersects W in a nonzero
vector? These subspaces cannot be orthogonal.
(Restatement: Suppose V is a p-dimensional subspace of Rn and that W is a q-dimensional
subspace of Rn . Moreover, suppose that V and W are orthogonal as subspaces of Rn . What
must be true of the quantity p + q?)
Answer: I claim that if p + q > n, then V and W intersect in some nonzero vector.
Equivalently, if V and W are orthogonal subspaces, then p + q ≤ n.
To see this, suppose p + q > n. Let {v1 , . . . , vp } be a basis for V and {w1 , . . . , wq } be a basis
for W. Then, since Rn is n-dimensional and since p + q > n, it must be the case that
{v1 , . . . , vp , w1 , . . . , wq }
is a linearly dependent set. Thus, there exist real numbers c1 , . . . , cp+q not all zero such that
c1 v1 + . . . + cp vp + cp+1 w1 + . . . + cp+q wq = 0.
Stated another way,
c1 v1 + . . . + cp vp = −cp+1 w1 − . . . − cp+q wq .
The vector on the left-hand side is clearly an element of V, whereas the vector on the right-
hand side is just as clearly an element of W. Therefore, the above vector (which is necessarily
nonzero) is an element of the intersection V ∩ W.

3
5. Problem 3.1.42. Suppose S is spanned by the vectors (1, 2, 2, 3) and (1, 3, 3, 2). Find two
vectors that span S⊥ . This is the same as solving Ax0o for which A?
Answer: Since the given vectors live in R4 and are linearly independent, we see that S is
2-dimensional. Thus, it makes sense that S⊥ should also be 2-dimensional.
Since we know that the orthogonal complement of the row space of any matrix is the nullspace
of the matrix, we can determine S⊥ simply by constructing a matrix A whose row space is S
and then determining the nullspace of A. Clearly, if
 
1 2 2 3
A=
1 3 3 2

then the row space of A is equal to S. To find the nullspace of A, we want to row reduce the
augmented matrix  
1 2 2 3 0
.
1 3 3 2 0
First, subtract row 1 from row 2:
 
1 2 2 3 0
.
0 1 1 −1 0

Next, subtract twice row 2 from row 1:

 
1 0 0 5 0
.
0 1 1 −1 0

The corresponding system of equations is

x1 + 5x4 = 0
x2 + x3 − x4 = 0

or, equivalently,

x1 = −5x4
x2 = −x3 + x4 .

Hence, the solutions to the equation Ax = 0 are of the form

   
0 −5
 −1   1 
x3 
 1  + x4  0  .
  

0 1

Thus,    

 0 −5 

−1   1
 
 1 , 0
   
 
 
0 1
 

is a basis for the nullspace of A, i.e. a basis for S⊥ .

4
6. Problem 3.1.46. Find AT A if the columns of A are unit vectors, all mutually perpendicular.
Answer: Suppose A is m × n, with columns a1 , . . . , an . Then

— aT1 — 
 T
a1 a1 aT1 a2 . . . aT1 an
  

 — aT —  | | |  aT a1 aT a2 . . . aT an 
2  2 2 2
AT A =  a a2 . . . an  =  . .
  
.. 1 . . ..
 .. .. ..

 . 
| | | . 
— aTn — aTn a1 aTn a2 . . . aTn an

In other words, the (i, j) entry of the matrix AT A is equal to

aTi aj = hai , aj i.

Since the columns are all mutually perpendicular, the above inner product will be zero unless
i = j, meaning that the only nonzero entries in AT A will be on the diagonal.
Moreover, hai , ai i = kai k2 = 1 since each column is a unit vector. Therefore,
 
1 0 ... 0
 0 1 ... 0 
AT A =  . . ,
 
 .. .. . . .. 
. . 
0 0 ... 1

the identity matrix.

7. Problem 3.2.8. The methane molecule CH4 is arranged as if the carbon atom were at the
center of a regular tetrahedron with four hydrogen atoms at the vertices. If vertices √ are
placed at (0, 0, 0), (1, 1, 0), (1, 0, 1), and (0, 1, 1)—note that all six edges have length 2,
so the tetrahedron
is regular—what is the cosine of the angle between the rays going from
the center 21 , 12 , 21 to the vertices? (The bond angle itself is about 109.5◦ , an old friend of
chemists).
Answer:

In the above figure, the black dot is the carbon atom and the blue segments are the bonds
between carbon and hydrogen atoms (the gray segments are the edges of the tetrahedron and

5
don’t mean anything chemically). The goal is to determine the cosine of the angle θ between
two of the blue segments.
The top left blue segment corresponds to the vector
   1   1 
1 2 2
v =  0  −  12  =  − 21  ,
1 1
1 2 2

whereas the top right blue segment corresponds to the vector

   1   1 
0 2 −2
w =  1  −  12  =  21  .
1 1
1 2 2

We know that
hv, wi = kvkkwk cos θ,
so
hv, wi
cos θ = .
kvkkwk
The numerator is
− 21
 
 
1 1 1   = −1 − 1 + 1 = −1.
hv, wi = vT w = − 1
2
2 2 2 1 4 4 4 4
2

To find the first term in the denominator, note that

 1 
 
1 1 1  21  1 1 1 3
kvk2 = hv, vi = vT v = − −2 = + + = ,
2 2 2 1 4 4 4 4
2

so r √
p 3 3
kvk = kvk2 = = .
4 2
A similar calculation shows that √
3
kwk =
2
as well, so
3
kvkkwk = .
4
Therefore, putting it all together,

hv, wi −1/4 1
cos θ = = =− .
kvkkwk 3/4 3

6
8. Problem 3.2.14. What matrix P projects every point of R3 onto the line of intersection of
the planes x + y + t = 0 and x − t = 0?
Answer: First, we need to find a vector on the line of intersection. Note that any such vector
is necessarily a solution of the matrix equation
   
  x 0
1 1 1  y  =  0 .
1 0 −1
t 0

To solve this equation, we’ll do Gaussian elimination on the augmented matrix

 
1 1 1 0
.
1 0 −1 0

Subtract row 1 from row 2 to get

 
1 1 1 0
,
0 −1 −2 0

Then add row 2 to row 1 and multiply row 2 by −1:

 
1 0 −1 0
.
0 1 2 0

This corresponds to the system of equations

x1 − x3 = 0
x2 + 2x3 = 0

or, equivalently,

x1 = x3
x2 = −2x3 .

Therefore, solutions to the matrix equation are of the form

 
1
x3  −2  ,
1
 
1
meaning that the vector a =  −2  is in the desired line of intersection.
1
Now, just recall that the projection matrix P is given by

a aT
P = .
aT a

7
The denominator is simply
 
1
aT a = [1 − 2 1]  −2  = 1 + 4 + 1 = 6.
1

The numerator is    
1 1 −2 1
a aT =  −2  [1 − 2 1] =  −2 4 −2  .
1 1 −2 1
Therefore,   1
− 13 1
 
1 −2 1
a aT1 6 6
P = T =  −2 4 −2  =  − 13 2
3 − 13  .
a a 6 1
1 −2 1 6 − 13 1
6

8
9. Problem 3.2.18. Draw the projection of b onto a and also compute it from p = x
ba:
       
cos θ 1 1 1
(a) b = and a = . (b) b = and a = .
sin θ 0 1 −1

Answer: (a)

p a

aT b
Since x
b= aT a
, we just compute
 
cos θ
aT b = [1 0] = cos θ
sin θ

and  
T 1
a a = [1 0] = 1,
0
so
aT b cos θ
x
b= T
= = cos θ.
a a 1
Therefore,    
1 cos θ
p=x
ba = cos θ = .
0 0

(b)

9
 aT b
Since x
b= aT a
, we just compute
 
1
aT b = [1 − 1] =1−1=0
1
and  
T 1
a a = [1 − 1] = 1 + 1 = 2,
−1
so
aT b 0
x
b= = = 0.
aT a 2
Therefore,    
1 0
p=x
ba = 0 = .
0 0

10. Problem 3.2.24. Project the vector b = (1, 1) onto the lines through a1 = (1, 0) and a2 =
(1, 2). Draw the projections p1 and p2 and add p1 + p2 . The projections do not add to b
because the a’s are not orthogonal.
Answer:
a2

p2
b

p1 + p2

p1 = a1

First, notice that the projection of b onto the line through a1 is

aT1 b
p1 = a.
aT1 a1
Since  
1
aT1 b = [1 0] =1+0=1
1
and  
1
aT1 a1 = [1 0] = 1 + 0 = 1,
0
we have that
aT b
   
1 1 1
p1 = T1 a1 = = .
a1 a1 1 0 0

10
Next, the projection of b onto the line through a2 is

aT2 b
p2 = a2 .
aT2 a2

Since  
1
aT2 b = [1 2] =1+2=3
1
and  
1
aT2 a2 = [1 2] = 1 + 4 = 5,
2
we have that
aT b
   
3 1 3/5
p2 = T2 a2 = = .
a2 a2 5 2 6/5

Therefore,      
1 3/5 8/5
p1 + p2 = + =
0 6/5 6/5

11